Sometimes a model is transpiled to a standard model language/object which can later be read and used via other languages depending on use cases.
Example tools
Orchestrator should be able to parallelize and sequentialize tasks according to their dependencies. For machine learning perspective, a task can be anything related to data, e.g., ETL, data sanity check, and model training.
Sequential tasks
[Task A] → [Task B] → [Task C]
Parallel tasks
[Task A] → [Task B] → [Task C]
[Task D] → [Task E]
Example tools
Docker wraps an application with its software requirements, e.g., runtime, libraries, system tools, and settings.
Deployment tool is used to move your code or packages between environment and ensures that those packages are identical along the deployment process.
Example tools
After a model is deployed, there are certain things required to be monitored:
- Engineering perspective: resource utilization, latency, etc.
- Data science perspective: data (concept) drift, model metric, missing data, etc.
See a good read on model monitoring here.
Example tools